Automated scoring of rehabilitative tests via time-frequency informed singular spectrum analysis

Continual assessment of rehabilitation progress is necessary to enhance the effectiveness of therapy. In a previous work [1] this has been addressed by looking into the eigenvalues of the arm movement signals using singular spectrum analysis (SSA). The method however ignores the effect of data nonstationarity which includes at least three different trajectory segments. In this paper, the above work is refined by separating the effective signal segments using time-frequency transform before applying the SSA. The automation of Action Research Arm Test (ARAT), which is widely used as a valid and effective test to monitor the rehabilitation effects, is the objective of this paper. The proposed SSA is informed by the time-frequency properties of the data, resulting in a more accurate selection of eigenvalues to aid in signal filtering. The improvement over our previous attempt is evident and very promising.